KR20210065228A - Ultrasonic Lesion Extraction Method Using Medical Image Big Data - Google Patents

Abstract

Disclosed is an ultrasound lesion extracting method using medical image big data, which comprises the steps of: setting, by an image acquisition module, a deep learning-based image algorithm for collecting a DICOM standard format or digital image and clinical data in accordance with a digital ultrasonography of a digital ultrasound device; setting, by an image inquiry module, a DICOM image stored in a database as a reading state after an image storage module stores the DICOM image in the database; extracting, by an image processing module, an ultrasound feature preset for the DICOM image corresponding to acquired patient information and advancing a recognition algorithm; advancing an image algorithm using image preprocessing for the DICOM image; and generating reference standard data for each age group for reading a liver ultrasound lesion in a DICOM analysis module. Accordingly, the reliability of a test result can be improved.

Description

Ultrasonic Lesion Extraction Method Using Medical Image Big Data

The present invention relates to a method for extracting ultrasound lesions using medical image big data, and more specifically, a multi-tenancy big data DB that guarantees indexes of 200,000 or more per second as an ultrasound medical image lesion extraction algorithm based on Open Source Software, and It relates to an ultrasound lesion extraction method using medical image big data that enables the development of an ultrasound lesion extraction algorithm and provides real-time advanced analysis and visualization services using open software.

In recent years, interest in growth has increased, and growth-related disorders are frequently occurring as growth-related diseases rapidly increase due to changes in eating habits during childhood and adolescence.

The medical imaging diagnostic system is a system for imaging equipment such as X-ray, CT, MRI, PET, endoscopy, capsule endoscopy, and ultrasonography, as well as software for image processing, analysis, and visualization of results. Research is ongoing.

Currently, there are problems in that an error range occurs and reliability is lowered because specialists (doctors) directly see and read the presence or absence of lesions using ultrasound images.

In addition, there is a problem in that a portion that is difficult for a reader (doctor) to visually read is found because a difference in the acquired images is greatly generated depending on the skill of a photographer in photographing an ultrasound image used for reading an ultrasound image.

In addition, if the reader is inexperienced in ultrasound imaging, the reliability of the reading result is low, so it is difficult for the reader to determine it by himself/herself.

It has problems in use in the field because of the problem that it is dependent on the reader's experience and skill and the reproducibility problem that the reading result varies depending on the reader's personal skill.

The present invention guarantees the reliability and reproducibility of ultrasound reading by developing an automated algorithm through ultrasound image image and mapping operation.

001) Republic of Korea Patent 10-2000-0005393 ‘Synthetic aperture focusing method of ultrasound image using flat wave’ (2000) 002) Republic of Korea Patent 10-2007-0124595 ‘Ultrasonic Synthetic Aperture Beamformer and Ultrasonic Imaging Device Using the Same’ (2007) 003) Korean Patent Registration 10-2008-0119625 ‘Ultrasound diagnosis method and system for generating synthetic Doppler spectral image’ (2008) 004) Korean Patent Registration 10-2008-0108452 ‘Method and System for Generating Ultrasound Image in Multi-Frequency Mode’ (2008) 005) Republic of Korea Patent No. 10-2009-0111746 ‘Ultrasound system and method providing ultrasound spatial synthesis image’ (2009) 006) Republic of Korea Patent 10-2011-0046121 'High-intensity connection ultrasonic generation device with close contact detection function' (2011) 007) Republic of Korea Patent 10-2011-0046125 'High-intensity connection ultrasonic generator with marking recognition function' (2011) 008) Korean Patent Registration 10-2016-0057469 ‘Feedback in Medical Ultrasound Imaging for High Intensity Focused Ultrasound’ (2016)

An object of the present invention was made to solve the problems of the prior art, and to automatically extract a region of interest independently from an input image photographed under various environmental conditions and accurately perform ultrasound within an error rate (±5%) range through image preprocessing. The goal is to provide an ultrasound lesion extraction method using medical image big data that maps to a reference image to realize reliability and reproducibility of ultrasound reading.

In order to achieve the above object, an ultrasound lesion extraction method using medical image big data according to the present invention,

setting, by the image acquisition module, a deep learning-based image algorithm that collects DICOM standard format or digital images and clinical data according to digital ultrasound imaging of digital ultrasound equipment; after the image storage module stores the DICOM image in a database, the image inquiry module sets the DICOM image stored in the database to a reading state; an ultrasound feature extraction and recognition algorithm advancement step preset for the DICOM image corresponding to the acquired patient information by the image processing module; upgrading an image algorithm using image preprocessing for the DICOM image; and generating reference standard data for each age group for reading the liver ultrasound lesion in the DICOM analysis module.

The step of setting the image algorithm may include: separating a partial image required for liver lesion estimation from an image stored in a medical image storage; Separating a partial image required for liver lesion estimation from an image stored in a medical image storage, generating and managing metadata for the image; extracting and classifying features of each image from sub-image data using artificial intelligence-based big data processing techniques; and filtering for noise and removing noise from the ultrasound image in order to apply a specific extraction algorithm.

The step of extracting and classifying the features of the image includes: extracting shape features for the sub-images separated by the ROI and using the entire area of the object to describe the shape based on the area; And to describe the model, using the main information expressing the outline of the object, normalizing the model model, and extracting the characteristics of the liver lesion; characterized in that it comprises a.

In order to extract and separate the region of interest, the extraction of the region of interest includes the steps of considering the characteristics of a lesion part, a medium state, cancer, and a tumor of an ultrasound image of the liver; and the region-of-interest separation algorithm is characterized by further comprising: separating an object by a contour defining an edge region of a tumor and a cancer mass.

In order to perform feature extraction and classification of the image, measuring similarity by detecting a texture feature for a sub-image separated by an ROI; and entropy, contrast, difference, homogeneity, standard deviation, average variance, etc. as texture parameters and classifying by applying image data including features.

According to another embodiment of the present invention, the ultrasound lesion extraction method using medical image big data through the liver lesion ultrasound identification algorithm specifies the contour and shape of the ultrasound image using the YOLO method, which is a deep learning-based FAST search technique. It is characterized in that it includes an ultrasound image learning process to ensure reproducibility like a specialist for tumors and liver lesions, and it is based on the subject's liver ultrasound image, but includes an epigastric ultrasound image reading technology.

The analysis data for the liver lesion ultrasound identification algorithm is calculated through a model designed to produce the same determination result as a specialist for liver lesions such as liver cancer and liver tumor, characterized in that the analysis data for the ultrasound image is generated. using a liver lesion analysis program using digital ultrasound image information; and including the de-identification of medical information and the big data analysis and visualization module and the subject's personal information encryption and medical image encryption algorithm LEA.

According to the present invention, it is possible to reduce errors in lesion judgment and provide a guide by accumulating accurate image data through the development of improved lesion reproducibility in ultrasound images. In addition, it has the effect of predicting the cancer incidence rate according to environmental changes by analyzing regional or individual data using the accumulated data.

In addition, according to the present invention, it is possible to prevent mistakes that a subject may miss through ultrasound image analysis and to improve the reliability of test results by developing an automatic reading algorithm.

In addition, according to the present invention, there is an effect of supporting doctors in diagnosing cancer and selecting a treatment method through data mining to find new information based on a large amount of data.

1 is a flowchart of software development for automatic ultrasound image lesion reading using big data.
2 is an explanatory diagram of an ultrasound image lesion automatic reading SW block using big data.
3 is a diagram illustrating an algorithm development method for determining a lesion for an epigastric ultrasound.

The present invention extracts a region of interest based on an image processing function, and the extracted region of interest performs a mapping operation with an enhanced ultrasound image through an intelligent algorithm to derive a result.

In order to accomplish the above object, a step of developing an automatic reading algorithm using an ultrasound image and a step of constructing a hand sub-image library are included.

In the development of automatic ultrasound reading software using the big data, DICOM standard format or digital images and clinical data of the livers of clinical participants are collected at the hospital and stored in the story to separate the partial images necessary for estimating liver cancer and liver tumors. Creates, stores, and manages metadata about

In order to quantify discontinuous changes in pixel values in image processing in the deep learning-based image mining step using the ultrasound image data, a segmented region technique using texture feature values is used. Morphological analysis using textural features is not a static process, but a process that is constantly evolving and changing, so morphology analysis is a suitable method for measuring liver ultrasound images.

The region-of-interest extraction considers abdominal ultrasound of liver cancer and liver tumor, and the region-of-interest separation algorithm separates the object by the contours appearing in the tumor and cancer of the ultrasound image.

The shape features are extracted for sub-images separated by ROI, and the area-based method uses the entire area of the object to describe the shape. Similarity is measured by detecting texture features for sub-images separated by ROI. As texture parameters, entropy, contrast, difference, homogeneity, standard deviation, average variance, etc. are calculated and image data including features are applied and classified.

Hereinafter, an ultrasound lesion extraction method using medical image big data according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart of automatic reading SW development for a liver ultrasound image using an ultrasound image according to an actual example.

First, we collect clinical data and DICOM standard format or digital images of livers of clinical participants in the hospital.

Next, the partial image required for lesion estimation is separated from the image stored in the medical image storage, and metadata about the image is created for storage management, and features of each image by using artificial intelligence-based big data processing techniques from the sub-image data. extract and classify them.

To extract and classify features of an image, the following method is used. The shape feature is extracted for sub-images separated by ROI, and area-based uses the entire area of the object to describe the shape. To describe the model, the main information expressing the outline of the object is used, the model is normalized, and the characteristics of the mandible are extracted. Similarity is measured by detecting texture features for sub-images separated by ROI. As texture parameters, entropy, contrast, difference, homogeneity, standard deviation, average variance, etc. are calculated and classified by applying image data including features.

An ultrasound lesion extraction method using medical image big data according to the present invention comprises the steps of: setting, by an image acquisition module, a deep learning-based image algorithm to collect digital images and clinical data in DICOM standard format or digital images according to digital ultrasound imaging of digital ultrasound equipment; after the image storage module stores the DICOM image in a database, the image inquiry module sets the DICOM image stored in the database to a reading state; an ultrasound feature extraction and recognition algorithm advancement step preset for the DICOM image corresponding to the acquired patient information by the image processing module; upgrading an image algorithm using image preprocessing for the DICOM image; and generating reference standard data for each age group for reading the liver ultrasound lesion in the DICOM analysis module.

The step of setting the image algorithm may include: separating a partial image required for liver lesion estimation from an image stored in a medical image storage; Separating a partial image required for liver lesion estimation from an image stored in a medical image storage, generating and managing metadata for the image; extracting and classifying features of each image from sub-image data using artificial intelligence-based big data processing techniques; and filtering for noise and removing noise from the ultrasound image in order to apply a specific extraction algorithm.

In order to extract and classify the features of the image, shape features are extracted for the sub-images separated by the ROI, and the entire area of the object may be used to describe the shape in the area-based method. In addition, in order to describe the model, it is possible to use the main information that expresses the outline of the object, to normalize the model model, and to extract the features of the liver lesion.

In addition, in order to extract and separate the region of interest, the region of interest extraction considers the lesion part of the ultrasound image of the liver region, the state of the medium, cancer, and the characteristics of the tumor, and the region of interest separation algorithm separates the tumor and the edge region of the cancer mass. Objects are separated by defined contours.

To extract and classify features, measure similarity by detecting texture features for sub-images separated by ROI, calculate entropy, contrast, difference, homogeneity, standard deviation, mean variance, etc. as texture parameters and apply image data including features to classify

In the method for developing an ultrasound identification algorithm for liver lesions, the contour and shape of the ultrasound image are specified using the YOLO method, which is a deep learning-based FAST search technique. Including, based on the ultrasound image of the subject's liver, but includes an epigastric ultrasound image reading technology.

Analysis data for the liver lesion ultrasound discrimination algorithm is calculated through a model designed to produce the same determination result as a specialist for liver lesions such as liver cancer and liver tumor, thereby generating analysis data for the ultrasound image.

3 is a method for developing an ultrasound lesion discrimination learning algorithm. In general, detection is performed according to a standard designed in consideration of the method of predicting from the contour of the ultrasound image, the brightness, and the density of the medium obtainable from the gamma image information.

At this time, when it is assumed that the density of the relevant site is relatively large compared to that of a normal ultrasound imaging medium, information on the relevant area shown in the image is analyzed to determine whether it is a normal organ tissue, a cancerous mass, or a tumor. Based on this data, the lesion can be obtained and the type of lesion can be known through image learning.

Although the present invention has been described in more detail with reference to several embodiments above, the present invention is not necessarily limited to these embodiments, and various modifications may be made within the scope without departing from the technical spirit of the present invention.

Claims (7)

setting, by the image acquisition module, a deep learning-based image algorithm that collects DICOM standard format or digital images and clinical data according to digital ultrasound imaging of digital ultrasound equipment;
after the image storage module stores the DICOM image in a database, the image inquiry module sets the DICOM image stored in the database to a reading state;
an ultrasound feature extraction and recognition algorithm advancement step preset for the DICOM image corresponding to the acquired patient information by the image processing module;
upgrading an image algorithm using image preprocessing for the DICOM image; and
Ultrasonic lesion extraction method using medical image big data, characterized in that it comprises; generating reference standard data for each age group for reading the liver ultrasound lesion in the DICOM analysis module. The method according to claim 1,
The step of setting the image algorithm is,
Separating a partial image required for liver lesion estimation from an image stored in a medical image storage;
Separating a partial image required for liver lesion estimation from an image stored in a medical image storage, generating and managing metadata for the image;
extracting and classifying features of each image from sub-image data using artificial intelligence-based big data processing techniques; and
An ultrasound lesion extraction method using medical image big data, comprising the steps of: filtering for noise and removing noise from an ultrasound image in order to apply a specific extraction algorithm. 3. The method according to claim 2,
The step of extracting and classifying the features of the image,
extracting shape features for the sub-images separated by the ROI and using the entire area of the object to describe the shape based on the area; and
Ultrasonic lesion extraction method using medical image big data, comprising the steps of using main information expressing the outline of an object to describe the model, normalizing the model model, and extracting the characteristics of the liver lesion. 3. The method according to claim 2,
In order to extract and separate the region of interest,
The extraction of the region of interest includes the steps of considering the characteristics of the lesion part of the ultrasound image of the liver region, the state of the medium, cancer, and the tumor; and
The region of interest separation algorithm further comprises the step of separating an object by a contour defining an edge region of a tumor and a cancer mass; 4. The method according to claim 3,
In order to perform feature extraction and classification of the image,
measuring similarity by detecting texture features for sub-images separated by ROI; and
An ultrasound lesion extraction method using medical image big data, comprising: calculating entropy, contrast, difference, homogeneity, standard deviation, mean variance, etc. as texture parameters and applying image data including features to classify. In the ultrasound lesion extraction method using medical image big data through the liver lesion ultrasound identification algorithm,
The liver lesion ultrasound discrimination algorithm,
The YOLO method, which is a deep learning-based FAST search technique, specifies the outline and shape of the ultrasound image, and includes the ultrasound image learning process to ensure the same reproducibility as a specialist for liver cancer, liver tumor, and liver lesion. Ultrasound lesion extraction method using medical image big data based on but including epigastric ultrasound image reading technology. 7. The method of claim 6,
The analysis data for the liver lesion ultrasound discrimination algorithm,
Using a liver lesion analysis program using digital ultrasound image information, characterized in that the analysis data for the ultrasound image is generated by calculating through a model designed to produce the same determination result as a specialist for liver lesions such as liver cancer and liver tumors. step; and
An ultrasound lesion extraction method using medical image big data, comprising: including a de-identification and big data analysis and visualization module of medical information, and an encryption algorithm LEA for personal information encryption and medical image of the subject.

Images